@inproceedings{harrison-etal-2019-maximizing,
    title = "Maximizing Stylistic Control and Semantic Accuracy in {NLG}: Personality Variation and Discourse Contrast",
    author = "Harrison, Vrindavan  and
      Reed, Lena  and
      Oraby, Shereen  and
      Walker, Marilyn",
    editor = "Balakrishnan, Anusha  and
      Demberg, Vera  and
      Khatri, Chandra  and
      Rastogi, Abhinav  and
      Scott, Donia  and
      Walker, Marilyn  and
      White, Michael",
    booktitle = "Proceedings of the 1st Workshop on Discourse Structure in Neural NLG",
    month = nov,
    year = "2019",
    address = "Tokyo, Japan",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/W19-8101",
    doi = "10.18653/v1/W19-8101",
    pages = "1--12",
    abstract = "Neural generation methods for task-oriented dialogue typically generate from a meaning representation that is populated using a database of domain information, such as a table of data describing a restaurant. While earlier work focused solely on the semantic fidelity of outputs, recent work has started to explore methods for controlling the style of the generated text while simultaneously achieving semantic accuracy. Here we experiment with two stylistic benchmark tasks, generating language that exhibits variation in personality, and generating discourse contrast. We report a huge performance improvement in both stylistic control and semantic accuracy over the state of the art on both of these benchmarks. We test several different models and show that putting stylistic conditioning in the decoder and eliminating the semantic re-ranker used in earlier models results in more than 15 points higher BLEU for Personality, with a reduction of semantic error to near zero. We also report an improvement from .75 to .81 in controlling contrast and a reduction in semantic error from 16{\%} to 2{\%}.",
}